Average recurrence plots as estimators of ensemble recurrence probability
Recurrence plots encode returns of a trajectory to neighborhoods of earlier states. Many experiments and simulations, however, produce an ensemble of realizations rather than a single trajectory. We formulate the Average Recurrence Plot (ARP) as the entrywise sample mean of recurrence indicators obtained from independent or exchangeable ensemble draws. ARP thereby estimates a two-time recurrence-probability kernel at a specified state representation, dissimilarity, and threshold. This formulation supplies three advances beyond visual averaging: a precise estimand, finite-replicate uncertainty reporting, and direct validation against analytically tractable stochastic processes. For stationary Gaussian processes, including Ornstein–Uhlenbeck, autoregressive, and periodically wrapped dynamics, the expected diagonal-band profile is determined by the lag-dependent increment distribution. We then show why ARP is particularly natural for delay differential equations, whose initial conditions are functions. In Mackey–Glass dynamics, single-realization recurrence plots vary with initial history whereas ARP retains reproducible periodic or irregular recurrence geometry; the associated τ -recurrence-rate profile identifies the time scales producing secondary diagonal lines. A compact delayed-feedback example further demonstrates how the same ensemble observable reports reorganization of return-time structure. The result is a replicate-aware recurrence observable that is interpretable, uncertainty-aware, and reproducible across stochastic and history-dependent dynamical ensembles.
Authors
- Chikoo Oosawa (ORCID: https://orcid.org/0009-0007-8824-5465)
Publication Details
- Journal
- Physical review. E
- Published
- 2026-11-19
- DOI
- https://doi.org/10.1103/5wyr-s62q
- Primary Topic
- Chaos control and synchronization
- Type
- article
- Field-Weighted Citation Impact
- 0.00